Research On Lifespan Prediction For Hemispherical Resonator Gyroscope

The hemispherical resonator gyroscope (HRG) is a new vibration gyro, which has excellentfeatures of high reliability, long lifespan, radiation hardening and high accuracy. So it is widelyapplied in aircraft navigation, communication satellite stabilization, precision pointing andplanetary exploration, et al. Because of its characteristics of high reliability, long lifespan, highcost and small sample, it is impracticable to predict HRG’s lifetime via the whole life test.Therefore, it is significant to evaluate HRG’s lifetime before being applied in those high valuetasks. To predict lifetime for8HRGs, the thesis’s work mainly contains several aspects as follows:Firstly, wavelet and envelope-wavelet analysis are respectively employed to reduce noises inHRG’s original drift data. Low-noise and more regular data are obtained with data preprocessing.And the results show that both methods can reduce noises, and envelope-wavelet analysis canachieve more complete and higher accuracy results than wavelet analysis.Secondly, we studied the widely used methods, such as grey model, back propagation neuralnetworks (BPNN), support vector machine (SVM) and their improved models. Both the partialdata and the whole data are employed to verify these models’ long-term prediction performance.The results indicate that neural network-based and support vector machine-based methods havepoor performance in long time response and they are not fit for long-term prediction, but the greytheory-based methods are. With the conclusions, GM(1,1) and its residual modified model areeutilized to forecast multi-period data sequences for8HRGs. With grey correlation analysismehtod, the life spans of8HRGs are predicted out, which are more than7years.Thirdly, with particle swarm optimization of BPNN (PSO-BPNN) and support vectorregression’s (SVR’s) excellent self-learning and self-adaption abilities as well as grey model’seffectiveness in long-term prediction, one novel prediction model is proposed: residual modifiedARGM(1,1)(redisual modified autoregressive GM(1,1)). With experimenting by using partial dataand whole data, the results show that residual modified ARGM(1,1) not only has goodself-adaption and long-term prediction ability, but also has higher prediction accuracy thanresidual modified GM(1,1).Fourthly, residual modified ARGM(1,1) model is applied to predict multi-period datasequences respectively with preprocessed data by using wavelet analysis and envelope-waveletanalysis. And then with grey correlation analysis method,8HRGs’ lifetimes are predicted out. Theexperimental results indicate that the lifespans gained with wavelet analyzed data are all longerthan13years, and those with envelope-wavelet analyzed data are over15years. Based on thelifespans of several oldest spacecraft in the world, the method proposed in the thesis is valid andthe predictive results also meet the long lifespan’s demand.In summary, the method in the thesis may be used as references for hemispherical resonatorgyro’s long lifespan research.